MATH 217C NOTES ARUN DEBRAY AUGUST 21, 2015

Transcription

1 MATH 217C NOTES ARUN DEBRAY AUGUST 21, 2015 These notes were taken in Stanford s Math 217c class in Winter 2015, taught by Eleny Ionel. I live-texed them using vim, and as such there may be typos; please send questions, comments, complaints, and corrections to a.debray@math.utexas.edu. Thanks to Ikshu Neithalath for finding a few errors. Contents 1. Introduction, Complex Manifolds and Holomorphic Maps: 1/6/ Complex Manifolds from Two Different Perspectives: 1/8/ Tensor Calculus for Complex Manifolds: 1/13/ Dolbeault Cohomology and Holomorphic Vector Bundles: 1/15/ Holomorphic Vector Bundles and More Dolbeault Cohomology: 1/20/ Hermitian Bundles and Connections: 1/22/ Kähler Metrics: 1/27/ Curvature on Kähler Manifolds: 1/29/ Hodge Theory: 2/3/ The Hard Lefschetz Theorem: 2/5/ Line Bundles and Chern Classes: 2/10/ /12/ Sheaves and Čech Cohomology: 2/17/ /19/ Special Holonomy and Kähler-Einstein Metrics: 2/24/15 29 A differential geometer whose work often uses the simplifications obtained by considering the complex domain explained to me that the additional structure of complex manifolds makes them more interesting, just as two sexes are more interesting than one, but various aspects of this argument are open to debate. Michael Spivak 1. Introduction, Complex Manifolds and Holomorphic Maps: 1/6/15 The audience of the class has quite a varied background; some days, some people will be quite bored and others will find it quite difficult. We ll try to stay at an elementary level; this is an introductory course in complex differential geometry, supposed to be a second-year graduate-level course. In particular, this is not a topics course. We re going to be most interested in complex manifolds and related structures, and we ll approach them with differential geometry and complex analysis. There s another approach which uses algebra, which we re not going to talk about as much. We re not going to follow any textbook particularly closely; here are the course topics (possibly out of order). Complex structures, almost complex structures, and integrability. Hermitian and Kähler metrics. Connections. Complex vector bundles, holomorphic vector bundles, connections and curvature. Chern classes and Chern-Weil theory. Many examples; in particular, the physicists will be interested in Calabi-Yau manifolds and Kähler-Einstein metrics and the relationship with physics. Cohomology theories: Hodge theory and Dolbeault cohomology. Vanishing theorems. Deformation theory and Kodaira-Spencer theory. 1

2 Since this is a second-year class, there will be no final and no midterm, but there will be homeworks; as usual, the best way to understand the material is to work through the examples. We mentioned that there are the two analytic or algebraic approaches to complex differential geometry; the best results follow from combining the two. Here are some references for the class, recommended but not followed exactly. Moroianu, Lectures on Kähler Geometry. These lecture notes are short and to the point, and thus serve as a useful introduction. The first six lectures cover elementary differential geometry, which is a useful reference, since it will be assumed in this class. Huybrechts, Complex Geometry. This is a more detailed reference, and uses more complex analysis, and a little algebra. This is a cleaned-up version of Griffiths and Harris Principles of Algebraic Geometry (which is nonetheless still mostly analytic). Voisin, Hodge Theory and Complex Algebraic Geometry: Volume I. This is less relevant to the course, but its algebraic approach is useful and interesting. D ly, Complex Analytic and Differential Geometry. This is also less relevant to the course; it mixes the analytic and algebraic approaches. Now, let us enter the world of mathematics. In this class, we want to study complex manifolds; we ll define them precisely in a moment, but one can think of these as spaces covered by charts, where each chart locally looks like C n, and the change-of-charts maps are holomorphic, i.e. complex differentiable; the linearization is complex linear. Complex manifolds are special cases of real manifolds, but there are real manifolds which cannot be complexified. Kähler manifolds are a special class of complex manifolds; not every complex manifold is Kähler. C n and CP n are Kähler, as are complex submanifolds of CP n and Calabi-Yau manifolds. These will have relatively nice properties. A real manifold may have several complex structures (which will play a role in deformation theory), and similarly a complex manifold may have several Kähler structures. This class will assume some knowledge of real differential geometry, and the study of real manifolds. One studies a real manifold by adding structure, e.g. a metric (turning it into a Riemannian manifold). Adding a metric isn t canonical, but forgetting it is. Similarly, we ll view a complex manifold as a complex structure on a real manifold, and we can add additional struture, such as a Hermitian metric (where the inner product is Hermitian), producing what is called a Hermitian manifold. Like a Riemannian metric, there is always such a metric and generally many choices, and forgetting the complex structure, one obtains a Riemannian metric. Kähler metrics (and their associated Kähler manifolds) are special cases of Hermitian manifolds. We ll develop tools to prove our main results, mostly from differential geometry (connections, bundles, curvature), but also a little from algebraic topology (typically cohomology). The idea is that a complex structure places restrictions on the underlying topology of the manifold. We will have three types of main results. (1) The Hodge and Lefschetz theorems impose strong restrictions on the cohomology of Kähler manifolds, and in particular implies that there are complex manifolds that do not admit a Kähler structure. (2) Vanishing results, e.g. the Kodaira vanishing theorem and its applications, such as Kodaira embedding. These results can also be stated in terms of cohomology (specifically, H 0 vanishes). Intuitively, this result states that if a complex line bundle has negative curvature, then it has no nontrivial holomorphic sections; then, the Kodaira vanishing theorem states that under the assumption of positive curvature, all of the higher cohomology groups H q (M, E) on the manifold M and line bundle E vanish (i.e. q > 1). With the Riemann-Roch theorem (which itself is a special case of an index theorem), we can reframe this as a criterion for nice Kähler manifold to be embedded in CP N for some large N. This is quite a surprise, because in the complex case things are very different from R n : there are no partitions of unity and no compact complex submanifolds of C n (other than a point), which ultimately follows from Liouville s theorem. (3) In deformation theory, one asks how many differentiable complex structures there are on a given smooth manifold. This is a bit of a hopeless question, but we can ask the infinitesimal version: given a complex structure, how any infinitesimal deformations are there? These end up being once again governed by another cohomology group, and in particular, complex structures come in finite-dimensional families, unlike, say, Riemannian structures. However, not all of these first-order variations are integrable (which yields global structures from local ones), but higher-order cohomology groups dictate whether a global structure can be obtained. A lot of questions concerning complex structure are difficult, and some are even open. For example, does S 6 admit a complex structure? We know it s not Kähler, but otherwise this is an open question. Note that S 6 is compact, but there s a possibility that it has a complex structure that can t be embeddied in C n. 2

3 Holomorphic maps and complex manifolds. Since everyone in the class has seen single-variable complex analysis, we can use it to move to several complex variables. Definition. A domain will refer to an open, connected subset of C n. For the time being, U will always refer to a domain. Definition. A function f : U C m is holomorphic if it is complex differentiable, i.e. df is complex linear. We ll clarify what complex linear means in just a little bit. This is equivalent to any of the following: f satisfies the Cauchy-Riemann equations. f = 0. f is analytic. f has a continuous complex differential. In the single-variable case, suppose U C and f : U C. Then, write w = f(z) and z = x + iy, so f can be thought of as a function into R 2 ; similarly, write w = u + iv. This means we can make the notion of complex linearity precise: let j : R 2 R 2 ( correspond ) to multiplication by i (well, just rotation by 90 ), so that j End(R 2 ). As a matrix, we 0 1 can write j =, so that j = id. This is a way of specifying complex structure on R 2, which we will see. With this identification of C and R 2, we can think of f : R 2 R 2 (or from a subset of R 2, and so on). As a real transformation, f has a real Jacobian ) f = J R f = ( u x v x corresponding to the differential df. Then, df is said to be complex linear if it commutes with j, i.e. df j = j df, and when the matrices are expanded out, this is equivalent to requiring that u x = v y and u y v y, u y = v x. This is an elliptic differential equation, and therefore its solutions have very strong regularity properties. This is part of the reason they re so well-behaved. Now, we can return to the case of several complex variables, and clarify what exactly it means to be holomorphic. The idea is that at each point, the linearization of a holomorphic function is complex linear, and these vary continuously. The standard complex structure on C n is given by multiplying by i in each coordinate. We have coordinates z = (z 1,..., z n ), and a Hermitian metric z = z z n 2. Each z k = x k + iy k, so we can write C n = R n R n, as z = (x, y). Then, we once again get j End(R 2n ) which multiplies by i in every coordinate. Specifically, ( ) 0 In j =, I n 0 which is a nice analogue of the single-variable case. Then, once again, complex linearity is equivalent to commuting with j. Holomorphic functions have some nice regularity properties. Let f : U C, where U C n. Then, df can be regarded as a C-valued 1-form n ω = (α k dx k + β k dy k ). k=1 But for the purposes of complex geometry, there s a better basis; instead of using dx k and dy k, it s better to use dz k and dz k, where z k = x k + iy k, so dz k = dx k + i dy k, and z k = x k iy k, so that dz k = dx k i dy k. In the specific case ω = df, the coefficients are familiar: df = ( f dx k + f dy k x k y k k One can (and should) verify that the dual basis to dz k and dz k is = 1 ( i ) z k 2 x k y k z k = 1 2 ( x k + i y k 3 ). ).

4 Mind the sign change. Then, we have operators and, defined as f = k f = k f z k dz k f z k dz k. Hence, df = f + f, and f is holomorphic iff f = 0. In this new basis, j is considerably easier to write down. Here are some useful properties of holomorphic functions. Most come from one variable, and many but not all extend to several variables; there are a few properties which are only true in several complex variables. Theorem 1.1 (Cauchy integral formula). Let U C be a ball (this can be made more general), f : U C be holomorphic on U so that it extends continuously to U. Then, for any z 0 U, f(z 0 ) = 1 f(z) dz. 2πi z z 0 From a differential-geometric perspective, this states that f(z) dz α = z z 0 is a closed one-form on U \ {z 0 } (which you can check by computing dα = 0), so this is essentially a corollary of Stokes theorem. This also relies on the calculation that 1 2πi z z 0 =ε U f(z) dz z z 0 ε 0 f(z 0 ). This generalizes to several complex variables thanks to Fubini s theorem. Let U be a polydisc (i.e. a product of discs), and f : U C n be holomorphic on U and extend continuously to U. Then, the iterated integral satisfies f(w) = 1 f(w) dw (2πi) n (w 1 z 1 ) (w n z n ), where dw = dw 1 dw n, where the integral is taken over the product of the boundaries. Regarding the integrand as a power series, we can see the following. Theorem 1.2 (Osgood). A holomorphic function of several variables is complex analytic (i.e. given by a complex power series). 2. Complex Manifolds from Two Different Perspectives: 1/8/15 The two perspectives will be thinking of complex manifolds as integrable almost complex manifolds (linear algebraic) and a more complex-analytic perspective. A complex manifold is just like a real manifold, except that the charts are subsets of C n and the transition maps are holomorphic. Definition. A n-dimensional complex manifold M is a smooth manifold on which there exists a collection of charts {(U α, ϕ α )} covering M, where each U α M is open and ϕ α : U α C n has open image. Furthermore, if α and β give two charts, then the transition maps ϕ 1 β ϕ α are holomorphic on the overlaps. The charts are called holomorphic charts, and the coordinates they induce are called holomorphic coordinates. Note. In this class, we will require all manifolds to be Hausdorff, and have a countable topological basis. Note. Just as the same topological manifold may admit many smooth structures, a smooth manifold can admit different complex structures; this will be useful in deformation theory. Definition. f : M C is holomorphic if f ϕ 1 α is holomorphic for all holomorphic charts. Similarly, f : M N is holomorphic if it is holomorphic as a map C m C n on each chart. One can check that these are well-defined notions, no matter which set of charts one uses. Note. There are no nonconstant holomorphic functions on a compact, connected, complex manifold, which is just a corollary of the maximum modulus principle (in several variables, which follows from the single-variable statement, by checking each coordinate). This is one of the huge differences between complex geometry and real geometry. 4

5 Definition. f : M N is a biholomorphism if it is a holomorphic bijection and f 1 is also holomorphic. This is the notion of equivalence in complex geometry, akin to a diffeomorphism in differential topology. If M and N are biholomorphic, one writes M N. Another interesting fact is that the size of the charts matters; Liouville s theorem (in one variable) implies that C D. Note. More worryingly, when n 2, the unit ball ( z < 1) is not holomorphic to the polydisc (the product of n discs, or z 1 < 1,..., z n < 1). The proof is not that easy, but ultimately it s because they have different automorphism groups; the automorphism group of the unit ball contains U(n), but that of the polydisc is abelian (once the origin is fixed). In particular, the Riemann mapping theorem does not extend to several variables (though the maximum modulus principle and analytic continuation both extend). This is why we will prefer polydiscs to balls for bases; it allows us more naturally to do complex geometry in each variable. Example 2.1. (1) C n is a complex manifold, with only one chart. More generally, any finite-dimensional complex vector space V is a complex manifold (which will be useful when there s no natural basis). In complex geometry, this is called affine space. (2) Similarly, any open subset of C n is also a manifold. (3) More nontrivially, we can make tori: consider a lattice Λ in C n, e.g. Λ = Z 2n, which can be thought of as n linearly independent vectors (over R) and then integer linear combinations of them. Then, C n /Λ is a complex manifold, and is diffeomorphic to T 2n = S 1 S }{{} 1. 2n times For example, if n = 1 and Λ = Z τz, for τ C \ R, the quotient is an elliptic curve. Not all elliptic curves are biholomorphic, and therefore not all tori of a given dimension are. E τ E τ iff τ = Aτ for an A SL(2, Z), i.e. the two lattices can be taken into each other by a linear transformation (which is where the special linear group comes from). Nonetheless, all tori of the same dimension are diffeomorphic. (4) We also have complex projective space CP n, which is geometrically the space of complex lines in C n+1. It s hard to put a complex structure on it from this perspective, so it s easier to think of it algebraically, as CP n = (C n+1 \{0})/C, where C = C\{0} and a λ C acts on C n+1 \{0} by (z 0,..., z n ) (λz 0,..., λz n ). Thus, points on the same line through the origin are identified, so this is really the same thing. Once we quotient out, keep the same coordinates: [z] = [z 0,..., z n ], where scaling factors are ignored (and brackets are used so we don t forget this). These are sometimes known as homogeneous coordinates. To get charts, take U i = {[z 0,..., z n ] CP n z i 0} and ϕ i : U i C n given by ϕ i ([z 0,..., z n ]) = ( z0 z i,..., z i 1 z i, z i+1 z i,..., z ) n. z i CP n \ U i is H i = {[z 0,..., z n ] z i = 0}, which is called the hyperplane at infinity in the i th coordinate. H i CP n 1, so H i is closed, and therefore U i is open. To see that ϕ i is a bijection, its inverse is just (w 0,..., w n ) [w 0,..., w i 1, 1, w i+1,..., w n ]. It s not too hard to check that the transition maps are holomorphic (because the product of holomorphic functions is holomorphic, and the quotient is too, whenever the denominator doesn t vanish, just as in the single-variable case). A special case of this is the Riemann sphere CP 1 = S 2. The two hyperplanes at infinity are the north and south poles. The process of obtaining CP n from C n+1 is called projectivizing; one can do this to any (finite-dimensional) complex vector space V to get P(V ) = (V \ {0})/C. This is of course biholomorphic to CP n, but there might not be a natural biholomorphism. (5) The complex Grassmanian of k-dimensional complex planes in C n is also a complex manifold; when k = 1 or k = n 1, this is the same as CP n 1. (6) The Hopf manifold is the quotient of C n \ {0} by Z acting as follows: fix a λ > 1 (the topology is independent of the choice of λ) and for k Z, k (z 1,..., z n ) = (λ k z 1,..., λ k z n ). The resulting quotient is a complex manifold diffeomorphic to S 1 S 2n 1, and will be important because when n 2, it will provide an example of a complex manifold that cannot admit a Kähler structure (for topological reasons). When n = 1, one gets a torus which is an elliptic curve, and it s possible to explicitly write down which lattice it corresponds to. 5

6 (7) To produce more examples, one can talk about complex submanifolds in precisely the same way as one would talk about real submanifolds. N is a k-dimensional complex submanifold of an n-dimensional manifold M if there exists a collection of holomorphic charts on M covering N such that the image of N under the chart maps is linear (specifically, C n k ). This definition is motivated by the inverse function theorem; there are plenty of other definitions, but fortunately they re all equivalent. Now, we can define lots of complex submanifolds of C n or CP n ; for example, if f : C n C is holomorphic, then the zero locus of f, i.e. Z(f) = f 1 (0), is a complex manifold by the implicit function theorem. Useful examples of this include polynomials; if f is cubic, then Sard s theorem is another way to show that for generic coefficients, the zero locus is smooth (i.e. 0 is a regular value) and therefore is a complex manifold. If f : C n+1 C is a homogeneous polynomial of degree k, i.e. f(λz 0,..., λz n ) = λ k f(z 0,..., z n ), e.g. f(z 0, z 1 ) = z z 0 z z 3 1, then Z(f) C n+1, and if 0 is a regular value (which is not always true), then the homogeneity guarantees that Z(f) descends as a submanifold of CP n. Now we can talk about varieties and all that. (8) Another large class is the complex Lie groups, groups with a complex manifold structure such that multiplication and inversion (equivalently (x, y) x 1 y) are holomorphic. For example GL(n, C) is a complex Lie group, but U(n) is not (for example, U(1) = S 1 isn t even-dimensional!), ultimately because all compact, complex Lie groups are abelian (because the adjoint map is holomorphic and bounded, and therefore constant), and then (a bit harder to show) one can show they re all tori. A lot of these examples were quotients; other examples exist, but they re sometimes harder to construct. Another Perspective. For the purposes of deformation theory, there s a different, equivalent definition of a complex manifold, and it will be useful for some other cases, too. This will be more linear-algebraic, involving differential calculus or tensor calculus on complex manifolds. The idea is to take the linear-algebraic notion of an almost complex structure and then require integrability. In the process of defining this, we can also discuss the Dolbeault cohomology. If M is a complex manifold, one would expect TM to be a complex vector bundle, and that s what we re going to see. Definition. An almost complex structure on a real manifold M is a vector bundle endomorphism J End(TM ) such that J 2 = id. Note. J makes TM into a complex vector bundle, as (a + ib) X = ax + bj X for all a, b R and X TM. Thus, J is really just multiplication by i. We ll define complex vector bundles later, but the general idea is that each fiber is a complex vector space and the trivialization functions can be chosen to be complex linear. Lemma 2.2. If M is a complex manifold, then multiplication by i in each holomorphic coordinate chart induces a well-defined almost complex structure on TM. Proof. We ll differentiate multiplication by i and see what happens, and then prove its independence of choice of holomorphic chart. Let s pick holomorphic coordinates (z 1,..., z n ), and decompose them into z k = x k + iy k, so we have basis vectors x k and y k for the tangent space. Then, define J as last lecture, i.e. J = y k and J = x k, and write z = (x, y), so C n = R n R n. Then, J is the same matrix as last time, and J 2 = id (as a matrix or just by the action on the basis), and it s independent of a change of coordinates, because a change of coordinates ψ is holomorphic, i.e. it commutes with J. For the rest of this lecture, assume J is an almost complex structure. A lot of the linear algebra we want still works for almost complex structures, though not all of it. Since J 2 = id, then its eigenvalues are ±i. After we complexify, it can even be diagonalized. Consider the complexified tangent space T C M = TM R C. Tensoring with C is what is meant by complexifying. T C M is a complex vector space of dimension n. Lemma 2.3. Let T 1,0 (M) (resp. T 0,1 (M)) denote the i (resp. i) eigenbundle (i.e. eigenspace at each fiber) of J. Then: (1) T C M = T 1,0 M T 0,1 M. (2) T 1,0 M = {X ij X X TM }. (3) T 0,1 M = {X + ij X X TM }. 6 ( x k ) ( y k )

7 Proof. Regard Z T C M as Z = X +iy for X, Y TM. Then, define T 1,0 M and T 0,1 M as in (2) and (3), respectively; one can check that for all Z T 1,0 M, J Z = iz, because J (X ij X) = J X ij 2 X = J X + ix = i(x ij X). Then, the decomposition (1) is immediate, because any X satisfies 2X = (X ij X) + (X + ij X). Next time, we ll talk about the integrability condition. 3. Tensor Calculus for Complex Manifolds: 1/13/15 Last time, we started talking about tensor calculus on the tangent bundle, but we can place it in the more general setting of complex vector spaces to make it useful in more places. Definition. A complex vector space is a real vector space V together with a real linear J : V V such that J 2 = id. This gives it the structure of a vector space over C, but this alternate definition is useful as well. Definition. The complex conjugage of a complex vector space (V, J ) is V = (V, J ). We can complexify V by taking V R C; it s already complex, but now we have the structure induced by C, so we can take its i-eigenspace V 1,0 and its i-eigenspace V 0,1 ; thus, V R C = V 1,0 V 0,1. As complex vector spaces, V 1,0 = V 0,1 = (V, J ) (where V 1,0 has complex structure given by multiplication by i). This extends to the dual V and exterior powers in the same way. Note. If V is already a complex vector space, then V R C = V V. (It also has a quaternionic structure, but that s less important.) Now, we can apply this linear algebra to manifolds. Assume (M, J ) is an almost complex manifold; then, we can apply the above to its tangent space: TM R C = T 1,0 M T 0,1 M. To use forms, we ll look at the cotangent bundle T M = Λ 1 M. We can complexify it, to obtain Λ 1 C M = Λ1,0 Λ 0,1. Definition. If M is a complex manifold with local holomorphic coordinates { (z 1,..., z n ). T 1,0 M is also denoted τm, and is called the holomorphic tangent space; its local basis is z k }. Similarly, T 0,1 M = τm is called the { anti-holomorphic tangent space, and has a local basis z k }. We want the dual structure to factor through the direct sum, so let Λ 1,0 = {ω Λ 1 CM ω(x) = 0 for all X T 0,1 M}, and Λ 0,1 is defined similarly (with T 1,0 in place of T 0,1 ). Thus, Λ 1,0 = {η iη J η Λ 1 M}, and Λ 0,1 = {η +iη J η Λ 1 M}. We can extend this to (p, q)-forms by using the fact that k Λ k (E F ) = Λ j E Λ k j F. j=1 Thus, we just wedge p things in Λ 1,0 and q things in Λ 0,1 together to get a basis, i.e. Λ p,q M = Λ p (Λ 1,0 M) Λ q (Λ 0,1 M). Then, Λ k C M = p+q=k Λp,q M. Definition. Denote by A p,q the space of smooth (p, q)-forms on M. That is, this is the space of sections Γ(Λ p,q ). In differential geometry, this space of sections is usually denoted Ω, but this will be used for holomorphic sections later. Again, if M is a complex manifold, we have some nice bases. A basis for Λ 1,0 is given by dz k = dx k + i dy k. Similarly, a basis for Λ 0,1 is given by dz k = dx k i dy k. If η A p,q (i.e. η is a (p, q)-form), then in local coordinates η = I,J η IJ dz I dz J, where dz i = dz i1 dz ip and dz J = dz j1 dz jq, where i 1 < < i p and j 1 < < j q. 1 The coefficients η IJ are smooth functions. 1 Sometimes, the notation ηij is used. This has no content, but is an interesting reminder, and sometimes is useful for keeping track of things. 7

8 All of this linear algebra works for almost complex manifolds, so let s talk about integrability, which will make the difference. Definition. Assume (M, J ) is an almost complex structure. Then, the Nijenhuis 2 tensor is Here, X, Y TM, though we can complexify. Exercise 3.1. Check that N J is actually a tensor. N J (X, Y ) = [X, Y ] + J [J X, Y ] + J [X, J Y ] [J X, J Y ]. The Nijenhuis tensor s signs aren t arbitrary, and we ll see how to derive them, Proposition 3.2. If J comes from a complex structure on M, then N J = 0. { This can be proven by checking for X, Y x k, y k }. The converse is also true, which is the harder direction. Theorem 3.3 (Newlander-Nirenberg Integrability Theorem). If N J = 0, then J comes from a complex structure on M. Proof sketch. If N J = 0 and J is real analytic, then the Frobenius integrability theorem tells us that J is integrable. But the hard part is the regularity result: it s beautiful, but hard to show that N J = 0 induces a differential equation which leads J to be real analytic. Proposition 3.4. Assume J is an almost complex structure on a smooth manifold M. Then, the following are equivalent: (1) J comes from a complex structure on M. (2) N J = 0. (3) T 1,0 M is formally integrable, i.e. for any Z, W T 1,0 M, [Z, W ] T 1,0 M. (4) T 0,1 is formally integrable. (5) d(a 1,0 ) A 2,0 A 1,1 (i.e. there s no (0, 2)-component). (6) d(a p,q ) A p+1,q A p,q+1. When we define, we will see that this is also equivalent to 2 = 0. Proof sketch. We re only going to show the easier parts, as we saw in the proof sketch of Theorem 3.3 it can get quite difficult. To see why (2) (3), suppose Z T 1,0 M, so that Z = X ij X for X TM. If we take X, Y to be local vector fields on M, then we can regard them as vector fields in T 1,0 M; then, let Z = X ij X and W = Y ij X, so one can check that [Z, W ] = N J (X, Y ) + ij N J (X, Y ). To see why (3) (5), take a (1, 0)-form ω, so that ω(z) = 0 for Z T 0,1 M. Thus, dω has no (0, 2)-part. This is equivalent to showing that dω(z, W ) = 0 for all Z, W T 0,1 M. But we can expand this out to dω(z, W ) = Z ω(w ) W ω(z) ω[z, W ], but the condition given shows ω(z) = ω(w ) = 0, so therefore dω has no (0, 2)-part iff the Lie bracket [Z, W ] T 0,1 M for all Z, W T 0,1 M, which is (4), and by complex conjugation, we get (3). Then, by the Leibniz rule, this can be extended to (6), and the other direction is immediate. What s left is the equivalences of (1), (2), and (3), which is more or less the content of Theorem 3.3 and uses the Frobenius integrability theorem. Now, let s use this. Proposition 3.5. Any almost complex structure on a real, orientable 2-manifold is integrable. Proof. This is by dimensionality: N J (X, X) = 0 and N J (X, J X) = 0, so that s all we need for a basis. Note. An almost complex structure is a topological condition (which we ll see because it involves cohomology), and therefore the only spheres that admit almost complex structures are S 2 and S 6. On S 6, one can explicitly create a nonintegrable almost complex structure by considering it to be the unit sphere in R 7, which can be regarded as the imaginary part of the octonions O (sometimes this is called the Cayley numbers). Then, there is a definition of a cross product p v p if v p, and then J p (v) = p v. Calabi showed this isn t integrable; there may be others which could be integrable, but this is still open. 2 Prononced nigh-house. 8

9 Condition (2) of Proposition 3.4 implies that if G is a complex Lie group, then its Lie algebra T e G is also a complex Lie algebra. One would expect (and it s true) that if M is a compact complex manifold, then Aut(M), the set of biholomorphic functions M M, is a Lie group. Its Lie algebra is the space of infinitesimal automorphisms of M, i.e. flows of holomorphic vector fields. We haven t defined these precisely; that s all right. Definition. If M is a complex manifold, then a holomorphic vector field on M is a vector field = X ij X T 1,0 M (i.e. X is real) such that the flow of X consists of holomorphic maps, i.e. L X J = 0. Later, we ll see that these are equivalent to holomorphic sections of τ M, the holomorphic vector bundle. For now, if (z 1,..., z k ) are holomorphic coordinates on M, then Z = k a k (z 1,..., z n ) z k, where the a k are holomorphic functions. Again, to show all of these things, there s a lot of linear algebra. Dolbeault cohomology. Suppose M is a complex manifold, so that condition (6) of Proposition 3.4 implies that we have a map d : A p,q A p+1,q A p,q+1. Thus, we can define = π p 1,q d and = π p,q+1 d. Thus, : A p,1 A p+1,q and : A p,q A p,q+1. These are both differential operators (i.e. C-linear and satisfying the Leibniz rule, which one can check); furthermore 2 = 2 = 0, and they anti-commute, as + = 0. This can be checked because d 2 = 0, and then expanding out. Now, we have something called the Hodge complex: A 0,0 and so on. A 1,0 A 0,1 A 2,0 A 1,1 A 0,2 Definition. The (p, q) Dolbeault cohomology group H p,q (M) is the (co)homology of A p, -closed (p, q)-forms on M mod the -exact (p, q)-forms on M. A p, +1. That is, it is the 4. Dolbeault Cohomology and Holomorphic Vector Bundles: 1/15/15 I don t know why, but I always forget the 1/(2πi) term in the Cauchy integral formula! Recall that we had a chain complex A p,0 A p,1, and the Dolbealut cohomology group H p,q (M) is the q th homology group of the complex. Definition. The Hodge numbers h p,q (M) of the manifold are given by the ranks of the respective Dolbeault cohomology groups. It s hard to calculate this from the definition, like any cohomology. Example 4.1. However, one simple case is H 0,0, because -closed (0, 0)-forms are holomorphic functions, and there are locally lots of them. In general, this could be infinite-dimensional, but if M is compact and connected, they re globally only constant, so H 0,0 (M) = C, which is nice. Why do we care about this construction? One useful case is the existence of an invariant of the complex structure. Another useful thing is that it s contravariant functorial: if f : M N is a holomorphic map of complex manifolds, then there s an induced f : H p,q (N) H p,q (M), and it is a group homomorphism. But Dolbeault cohomology is also useful in local deformation or obstruction theory, which can be useful for keeping track of obstructions to the existence of a complex structure on a manifold. Suppose one wants to solve the equation β = α for a (p, q)-form α. If this were to have a solution, α = 0 (i.e. α is -closed), which can be checked locally. If there is no obstruction, then the equation has local solutions, but we may not be able to patch them together to get a global solution on M. The global obstruction is governed by Hodge theory, which says that if the cohomology class [α] H p,q (M) vanishes, then we can do this. There are two powerful tools to deal with this. 9

10 Sheaf theory adopts the approach of looking at locally holomorphic functions and forms, and spends time worrying about how they patch together, which sheaves help with. A more analytic approach is to use Sobolev spaces, which relax the holomoprhic condition, but require one to worry about regularity. Ideally, one could use both, but one may be more useful than the other in a given situation. This lemma is really due to Dolbeault, but has Poincaré s name on it for some reason. Lemma 4.2 ( -Poincaré). Any -closed (p, q)-form is locally -exact. As a corollary, if P is the polydisc, then H p,q (P ) = 0 if q > 0. Proof sketch. For the complete proof, see Huybrechts book, 1.3. We ll use the polydisc for working locally, since analysis is easier in compact spaces. In one variable, assume g is smooth on the closed unit disc D; then, we want an f such that f = g dz. We can explicitly construct a solution: f(z) = 1 2πi D g(w) dw dw, w z which is smooth on D and satisfies f = g dz. This was a version of the Cauchy integral formula. Now, let s talk about (0, 1)-forms in n variables; let α = K α j dz j, k=1 where K n. We ll induct on K; since α is -closed, then each α j is holomorphic in z K+1,..., z n. One can integrate (made more precise in the book) the coefficient α K of dz K to get a β such that α β is a linear combination of dz 1,..., dz K 1, which is the necessary inductive step. Then, we can generalize to (0, q)-forms from (0, 1)-forms. Let A p,q K be space of (p, q)-forms that only involve dz 1,..., dz K. Then, filter A 0,q by A 0,q K ; one can show that if α 0 mod A0,q K and α is -closed, then one can find a β such that α β 0 mod A 0,q K 1. Finally, generalizing yet further to (p, q)-forms is immediate from (0, q), since we re merely adding dz terms. For some purposes, it may be easier to have cohomology vanish on the open polydisc, which can be analyzed by exhausting it by closed polydiscs that approach it in the limit. Thus, there s an extension problem, of taking a solution on one polydisc and extending it to a new solution on a larger polydisc which agrees (or mostly agrees) on the smaller one. Often in PDE theory, one considers a differential equation Df(x) = g(x) (where D is a differential operator). One way to solve this (intuitively, though it can be made precise) is to try to find a kernel k(x, y) which satisfies D x k(x, y) = δ x,y. 3 Then, general principles tell us that f(x) = k(x, y)g(y) dy is a solution of the differential equation. There are lots of important issues of convergence and making sense of this, but in the context where this works, D x f = D x k(x, y)g(y) dy = δ xy g(y) dy = g(x). This applies to complex geometry: for us, D =, and we want to solve f = g. Our kernel was the Cauchy kernel k(z, w) = dw 2πi(z w). Well, this is a one-form and not a function, and we re secretly using a metric, and so on; this isn t rigorous yet, but it can be and will be; this is the intuition for later on. Lemma 4.3 (Local Lemma). Let M be a complex manifold and ω be a real (1, 1)-form. Then, dω = 0 iff ω is locally of the form ω = i u for a (locally defined) real-valued function u. In this case, u is called the potential. 3 The δ function is not a function; it s a distribution, of course, but the equation can be reformulated such that this is all right. 10

11 Proof. Suppose we know that ω = i u, and we want to show that it s closed. Well, since ω is a (1, 1)-form, dω = ( + )( u) = 2 u + u = ( ) 2 u = 0, because they anticommute. The other direction uses both the real and complex versions of the Poincaré lemma; the real one is similar to the complex one, but follows from Stokes theorem. Assume ω is d-closed, so that dω = 0. Then. by the real Poincaré lemma, ω = dτ, for some real-valued 1-form τ. Then, thinking of τ as complex, we can uniquely decompose it as τ = τ 1,0 + τ 0,1 ; since τ is real, then τ 0,1 = τ 1,0 (which follows from the formula). Thus, ω = dτ = τ }{{ 1,0 } + τ 1,0 + τ 0,1 + }{{}} τ {{ 0,1 }. (2,0) (1,1) (0,2) But since this is a unique decomposition and ω is a (1, 1)-form, then τ 1,0 = 0, and similarly τ 0,1 = 0 too; thus, ω = τ 1,0 + τ 0,1. By Lemma 4.2, τ 0,1 = f for a (local) complex-valued function f, and similarly the complex conjugate τ 0,1 = f. Thus, so let u = Im(f). ω = τ 1,0 + τ 0,1 = f + f = 2 (i Im(f)), This proof follows a common approach of doing one thing from analysis followed by a bunch of linear algebra. We ve talked about holomorphic functions, so it may also be useful to discuss holomorphic forms as well. Definition. A holomorphic p-form is a (p, 0)-form ω such that ω = 0. Equivalently, ω is a holomorphic p-form iff it can be written as ω = ω I dz I, where the ω I are holomorphic functions. It turns out this will also be equivalent to ω being a section in a holomorphic vector bundle. Note that if q 0, a (p, q)-form α such that α = 0 isn t called holomorphic; we l see why in a bit. Holomorphic Vector Bundles. Intuitively, a holomorphic vector bundle is a complex vector bundle such that the transition functions are holomorphic. Equivalently, it is a complex vector bundle with a operator such that 2 = 0. This last condition means that Dolbeault cohomology only makes sense over holomorphic vector bundles. While one can define real and complex vector bundles over any topological space, holomorphic vector bundles can only exist over complex manifolds. Definition. Let M be a complex manifold. Then, a holomorphic vector bundle of rank r is a complex vector bundle π : E M such that the following hold. There exist holomorphic charts U α covering M and holomorphic local trivializations, i.e. the following diagram commutes. π 1 ψ α (U α ) U α C r π U α The transition functions g αβ : U α U β GL(n, C) are holomorphic. These are given as follows: going from β α, we have for x M and e C r, ψ α ψ 1 β (x, e) = (x, g αβ(x)e). Example 4.4. Let M be a complex manifold. The trivial bundle M C r is of course a holomorphic vector bundle. τm = T 1,0 M (i.e. involving just z ) is a holomorphic vector bundle (but T 0,1 M isn t). Its dual (τm) = Λ 1,0 M is holomorphic as well (which follows from its functorial properties); in complex and algebraic geometry, this is often denoted Ω M, called the bundle of holomorphic one-forms. Ω p = Λ p,0 M, the space of (p, 0)-holomorphic forms. 4 pr 1 4 These forms are sections of the bundle, so we should speak carefully, but they are frequently talked about as the same spaces. A bundle can also be thought of as a sheaf of sections. 11

12 The top power is a holomorphic line bundle, called the canonical line bundle. It has the special notation Ω n M = Λn,0 M. These are the analogues of volume forms, but in complex geometry. Here s why τm is holomorphic: choose local holomorphic coordinates (z 1,..., z n ) on M, so the local trivialization of τm has as a basis { z k }. We need to check that the change-of-coordinates and transition functions are holomorphic. If z = z(w) is a change of holomorphic coordinates, then w l = k z k, w l z k with no extra terms because w is also holomorphic; thus, these coordinate changes are holomorphic. Similarly, a trivialization of Ω p M has basis {dz I } I, where I = p for multi-indices I; one has to check that dw K = w K dz I z I has holomorphic functions (which relates to the Jacobian of the change of coordinates z w). I =p 5. Holomorphic Vector Bundles and More Dolbeault Cohomology: 1/20/15 You may have seen something like this [tautological line bundle] at IKEA. Recall that if M is a complex manifold, we have the holomorphic vector bundles, sections Ω M = τm = Λ 1,0 M, where τm = T 1,0 M. Then, Ω p M = Λp,0 M. Be careful, though; these are generally not Λ p,q for q > 0. A general principle here is tht most natural constructions on real vector bundles still work in the complex or holomorphic case. Here are some examples which stay holomorphic, where E and F are holomorphic vector bundles. The direct sum, E F. The tensor product E F. The dual E. The exterior powers Λ k E, and as a special case, the determinant line bundle Λ r E, where r is the rank of E. The pullback: if f : V M is holomorphic and E M is a vector bundle, then f : V N is a holomorphic vector bundle. One useful example is that if V N is a complex submanifold, the restriction of E V is defined as the pullback under inclusion i : V M, and is holomorphic. 5 Another useful fact about holomorphic vector bundles is that the total space is a complex manifold, with the projection holomorphic. Definition. Suppose V M is a complex submanifold. Let TM V denote the restriction of TM to V ; then, we have a short exact sequence 0 TV ϕ TM V N V 0, for a holomorphic bundle N V = coker(ϕ) called the normal bundle of V. Keep in mind that while short exact sequences of complex vector bundles always split (i.e. if 0 E F G 0 is short exact, then F = E G), this is not true for holomorphic bundles. For example, there s a rank 2 bundle E over an elliptic curve which fits into a short exact sequence 0 C E C 0 (where C denotes the trivial line bundle), but does not split. Since this sequence splits as a sequence of complex line bundles, we end up seeing that the same complex manifold may have nonisomorphic holomorphic structures. Nonetheless, the determinant behaves well in short exact sequences: if 0 E F G 0 is short exact, then det F = det E det G as holomorphic line bundles. In particular, we have the adjunction formula det(tm V ) = det TV det N V, which is used all the time. Example 5.1. Consider the tautological line bundle τ over CP n. Geometrically, we want the fiber of τ at a line l CP n to be the line l. Algebraically, consider τ to be the total space τ = {(l, u) CP n C n+1 u l}, which comes with the projection π onto the first factor. This is a holomorphic line bundle: consider coordinate charts U i on CP n, so z i 0, which comes with the projection π onto the first factor. This is a holomorphic line bundle: consider coordinate charts U i on CP n, so z i 0. The trivialization will be ψ i ([z], u) = ([z], u i ), which (you can check) plays well with transition functions g ij ([z])t = z i /z j, which is clearly holomorphic. When n = 1, CP 1 is the Riemann sphere C { }. Thus, we have two charts, z C and w = 1/z, whose overlap is C (one ignores 0, the other ignores ). Thus, the transition functions are g(w)(t) = w 1 t, for w C. 5 One can also pull back under the constant map, but the result is the trivial bundle. 12

13 This seems a little silly, but more interesting constructions, e.g. the canonical bundle or the tangent bundle, can be described in terms of τ (e.g. T M = τ 2 : the transitions of T M are between dz and dw, so dw = d(1/z) = 1/z 2 dz). Remark. A complex or holomorphic vector bundle is determined by the transitions functions g αβ : U α U β GL(r, C) (where r is the rank of the vector bundle), such that g αβ g βγ = g αγ on U α U β U γ, for all charts α, β, and γ. This is known as the Čech cocycle condition, and relates to Čech cohomology, which we ll discuss later. Dolbeault Cohomology with Values in a Holomorphic Vector Bundle. Let E M be a rank-r holomorphic vector bundle over an n-dimensional manifold. Choose a holomorphic trivialization (a local basis for the fiber) {e k } k=1,...,r, and let Λ p,q (E) = Λ p,q M E. Then, we can define ( n ) E ω k e k = k=1 k ( ω k ) e k, where ω k are (p, q)-forms, i.e. in holomorphic coordinates (z 1,..., z n ), ω k = I,J k=1 ω IJ k dz I dz J. Since 2 = 0, then 2 E = 0 as well, of course. And we can do some of the same stuff as before, taking smooth sections A p,q (E) of Λ p,q (E) (since we can only differentiate sections, such as vector fields, rather than vectors themselves), and we get maps A p,0 (E) E A p,1 (E) E A p,2 (E) E (1) This should look very familiar, and if E is the trivial line bundle we recover the Dolbeault cohomology from last time. Definition. The Dolbeault cohomology of M with values in E H p,q (M; E) is the homology of the complex (1), i.e. the E -closed (p, q)-forms with values in E modulo the E -exact (p, q)-forms. As a special case, H 0,0 (E) (sometimes Ω(E) or H 0 (M; E); the latter notation comes from Čech cohomology) is the sections σ of E which are holomorphic and satisfy E σ = 0, i.e. σ = k σ k (z) e k in a holomorphic frame, were the coefficients σ k are holomorphic functions on M. Additionally, the tautological vector bundle τ CP n has no holomorphic sections (other than the zero section), which is a little bit of a chore to calculate. There are more interesting examples, such as that of τ ; one appears in the homework. Remark. By the -Poincaré lemma (which still works just as well in this setup), over a polydisc (i.e. locally), (1) has trivial homology except in degree 0, where one gets Ω(E) = H 0,0 (M; E). This is called an affine or an acyclic resolution, and leads to the following theorem. Theorem 5.2 (Dolbeault). If H q (M, F ) denotes the q th Čech cohomology with coefficients in F, then H p,q (M; E) = H q (M, Ω p M E). This will be a little less mysterious once we actually define Čech cohomology. Integrability of Complex Vector Bundles. The question is, when is a complex vector bundle (over a complex manifold) a holomorphic vector bundle? The answer will be when there is an E operating on smooth sections of E, i.e. E : Λ 0,0 (E) Λ 0,1 (E) that acts as a differential operator. That is, it satisfies the Leibniz rule: when ω is a (p, q)-form and σ is a section of E, E (ω σ) = ( ω) σ + ( 1) q ω E σ. Furthermore, we require 2 E = 0. Once this exists, it can be extended using complex linearity and the Leibniz rule to E : Λ p,q (E) Λ p,q+1 (E). Theorem 5.3. Assume E M is a complex vector bundle over a complex manifold M. Then, a holomorphic structure on E is uniquely determine by a C-linear differential operator E : Λ 0,0 (E) Λ 0,1 (E) such that 2 E = 0. This should be thought of as a linear version of Theorem 3.3, but for vector bundles. 13

14 Proof sketch. All we have to do is find holomorphic trivializations to recover the holomorphic vector bundle, so let s do that. Notice that 2 E = 0 iff we can locally find a basis of E consisting of holomorphic sections of E (there s a little more to say here, but it s not as complicated as it looks). Now, we want to get local trivializations out of this. Going from a holomorphic bundle to this is relatively easy, so let s focus on the more difficult direction; suppose E is a complex vector bundle with a local basis {σ k } of smooth sections of E. Here are two things we know: that 2 E = 0, and n E σ k = τ kj σ j, (2) where E σ = τ σ. Here, σ = (σ 1,..., σ n ) in this basis, and τ = (τ ij ) is a matrix of (0, 1)-forms. Applying the Leibniz rule to (2), we get that since 2 Eσ k = 0, then Expanding out τ = (τ jk ), this also looks like j=1 τ + τ τ = 0. τ kl + j τ kj τ jl = 0. Now, we want to find a change of basis f : U GL(r, C) from our basis {σ k } to a holomorphic basis f = (f kl ). Let e k = f kj σ k ; we will want E e k = 0, i.e. f + f τ = 0. This is a PDE, and locally the obstruction to solving it turns out to be precisely the requirement that τ + τ τ = 0. 6 This last part is a little confusing (what is everything exactly?), but when M is one-dimensional, a lot of these just become functions, and the PDE has no obstruction; it s quite easy to integrate. In particular, (log f) + τ = 0, which is exactly the local Poincaré lemma. In higher dimensions, one looks for a local solution f = exp(f ) (the Lie group exponential), where F : U gl(r, C) (i.e. the Lie algebra), such that F + τ = 0. This is again the Poincaré lemma, with values in the Lie algebra, since in some vague sense log(f) = F. 6. Hermitian Bundles and Connections: 1/22/15 You shouldn t trust any of the signs I write... also any factors of 2πi. Today, we ll add extra data in the form of a connection and a metric, and will try to make it compatible with the complex structure and a bundle. There s an obstruction, which vanishes when the bundle is holomorphic, leading to a unique connection called the Chern connection, which is analogous to the Levi-Civita connection (especially when the given bundle is the tangent bundle). There will be a lot of different structures here, so be careful with exactly what goes where. Connections. Let E M be a real or complex vector bundle over a real manifold M. Definition. A connection is a linear differential operator : Γ(E) Λ 1 (E), i.e. from smooth sections 7 to one-forms on E (i.e. Λ 1 (E) = Γ(Λ 1 M E)), that satisfies the Leibniz rule (f σ) = df σ + f σ. Note that the linearity is real if E is real, and complex if E is complex. The Leibniz rule means that if v TM, then the directional or covariant or covariant derivative v σ corresponds to the usual notion of derivative. We can extend to p-forms with values in E, i.e. in Λ p (E), again by the Leibniz rule: if η Λ p (M) and σ Γ(E), then (η σ) = dη σ + ( 1) p η σ. Define the curvature operator R = 2, or more explicitly R σ = ( σ). This is a map from sections to sections, a 2-form with values in End(E). But we should do this in coordinates, so that we can actually calculate stuff. 6 This is technically a slightly illegal shortcut, since it s not always true that τ τ = 0. This is true in many cases, e.g. Riemann surfaces, line bundles, etc. If this is not true, often τ is real analytic, so it can be solved by the Frobenius theorem, expanding power series, and so on. Alternatively, we seem to have a twisted given by ( + τ)f = 0; one needs a gauge transformation to convert this to the standard. 7 Eventually, we ll have to weaken this regularity, and it will be important how many times things are differentiated. 14